Understanding public health compliance in emergencies

An investigation of the heterogeneity and pathways of factors that shape our behaviour during public health crises in LMICs

Jonas Weinert

March 5, 2025

Why

C19 produced a lot of research into pandemic behaviours and their determinants

But there is a substantive lack of insights into

  • How these relate to and interact with each other
  • Which factors have the strongest predictive value
  • LMIC specific dynamics
  • Causal links esp in LMICs

Outline

This thesis

Understanding the pathways of determinants of behaviour during pandemics in LMICs

Chapter Description
Lit review Overview of known asssociations with behaviour formation for NPI & vaccine uptake
Theory testing and development Comparative modelling of existing frameworks and lit infoirmed extensions
Health vs. Income Causal test and quantification of health vs income trade-off assumption
Peer effects Testing the effect of non self-selected peer groups on behaviour

Today’s presentation

  • Factors associated with pandemic behaviour formation
  • Theoretical Frameworks that (don’t) connect them
  • Comparative structural modelling to inform a theory of pandemic behaviour
  • Income vs health preservation
  • Peer effects vs self selection

Factors associated with pandemic behaviour formation

  • Financial constraints/ economic factors
  • Health risk perceptions
  • Sociological factors
  • Knowledge and Trust
  • Cultural factors
  • Institutional & political factors

Financial constraints/ economic factors

  • Classical economic theory suggests that individuals make rational choices under constraints, weighing the risk of infection against economic consequences (e.g., staying home vs. working).
  • Health capital model (Grossman, 1972): Health is treated as an investment, with lower-income individuals discounting future health risks if avoiding them means immediate economic hardship.
  • Prospect theory (Kahneman & Tversky, 1979): Individuals overweight immediate economic losses, making financially insecure people more likely to work despite health risks.

Empirical evidence

EG (Chetty et al., 2020; Papageorge et al., 2021) show that lower-income individuals and workers in precarious jobs were less likely to comply with lockdowns.

Health risk perceptions

  • Health Belief Model (Rosenstock, 1974): Compliance is influenced by perceived severity, susceptibility, and benefits of protective actions.
  • Risk perception (Slovic, 1987): Individuals facing immediate health threats (e.g., seeing loved ones infected) prioritize health over economic concerns.

Empirical evidence

E.G. Bartscher et al. (2021) show local infection rates predict compliance better than economic vulnerability; Allcott et al. (2020) find greater COVID-19 information exposure leads to higher adherence.

Sociological factors

  • Social capital theory: Stronger community networks lead to more collective action; U.S. counties with higher social capital had lower infection rates.
  • Social influence theory: Individuals conform to peer behavior, sometimes overriding economic incentives, with political affiliations and community norms influencing compliance.

Empirical evidence

E.G. Feelings of mutual obligation encourage compliance, with high-cohesion communities showing greater adherence to health measures (Zimmermann et al., 2022).

Knowledge and Trust

  • The “knowledge-behavior gap” occurs when trust and biases interfere with knowledge-driven compliance.
  • Trust in institutions: Compliance is higher when people trust expert sources, while peer/social media influence can sometimes outweigh expert advice.

Empirical evidence

Source of information matters: Walter & Xue (2024) find individuals believe information aligning with pre-existing views. Accurate knowledge helps individuals adopt appropriate behaviors (Baumgartner, 2023; Fadel et al., 2021).

Cultural factors

  • Religion
  • Etc

Institutional & political factors

Institutional trust (Fukuyama, 1995): High-trust societies show higher compliance, independent of financial constraints or personal risk perceptions.

Empirical evidence

Government policy & enforcement: Countries with higher institutional trust exhibited greater compliance across income levels.

Theoretical Frameworks that (don’t) connect these factors

Frameworks: Theory of plnned behaviour

REFERENCE/SOURCE

Frameworks: Health Belief Model

REFERENCE/SOURCE

Chapter: Comparative structural modelling to inform a theory of pandemic behaviour

What: Comparing and Extending TPB & HBM

  • Some studies used HBM OR TPB to analyse survey data
  • Only REFERENCE in LMIC (Looking at students in Turkey)
  • No paper integrating/ extending these in C19 in LMIC
  • No study comparing different geographical contexts

Where: Global comparison

Comparing contexts across Africa, Latin America, Asia

How: Structural Equation Modelling I

Structural Equation Modelling

  • Statistical technique that models complex relationships between observed (measured) and latent (unobserved) variables
    • Allows simultaneous estimation of direct and indirect effects in predicting preventive behavior
  • Direct, indirect, and total effects analysed for all models
  • Conducted at different waves for robustness checks and comparison

How: Structural Equation Modelling II

Confirmatory Factor Analysis (CFA)

  • To validate latent constructs
  • To operationalise constructs from survey items

Comparative Model Evaluation

  • Predictive power assessment (variance explained in preventative behavior)
  • Multi-group SEM for cross-country/SES/gender comparisons
  • Akaike Information Criterion (AIC) & Bayesian Information Criterion (BIC) for model comparison

With: Meta/MIT C19 preventative health survey

Data

  • Conducted via online surveys distributed to Facebook users globally
  • Over 100k observations across 78 countries
  • Weighted to account for probability of being FB user
  • Contains data on all aforementioned factors
  • Repeated cross sectional survey

Why: Contributions

Emprical contribution

  • First large scale SEM across countries and waves
  • Qunatification of uncertainity and relative importance of factors

Theoretical Contribution

  • Informing more comprehensive framework of pandemic behaviour

Policy relevance

  • Informs more efficient data collections during early phases of emergencies
  • Informs ressource allocation efficiency

Chapter: Income vs health preservation

What: Testing the trade-off assumption

Widely assumed trade-off that individuals face

  • Rationale behind cash transfer programmes
  • Supported by economic theory
    • Grossman (19X)
    • Prospect Theory
  • Empirical eveidence shows
    • Effects of income
    • Effects of risk perceptions
  • Income and risk perception never modelled together in causal way

What II: Hypotheses

H1(2): Economic vulnerability (health risk perceptions) predict compliance.

H3: Individuals face a trade-off between preserving their health and their economic security.

Where: Malawi during waves 1 & 2 of C19

How: Challenges of traditional econometric approaches I

How: Challenges of traditional econometric approaches II

Strengths

  • Interpretability
  • Causal pathway identification

Limitations

  • Exogeneity concerns
    • Either RCT or
    • Quasiexperimental approach difficult with income
    • Controlling requires many functional assumptions

How: Challenges of machine learning

Strengths

  • Very flexible re functional forms
  • Can handle large amounts of covariates

Limitations

  • Overfitting concerns
  • Causal pathway interpretation challenging

How: Double Debiased Machine Learning

Flexible ML to model covariate effect on treatment & outcome

  • Predict outcome/ treatment seperately given vector of covariates using ML methods

Orthogonalised regression on residuals to estimate causal effect of treatment and outcome

  • leftover variation of treatemnt -> leftover variation of outcome
    • Given no OVB
  • Based on Neyman (19XX)

How: Double Debiased Machine Learning II

Cross-fitting to reduce overfitting and regularization bias

With: Malawi High Frequency Phone Survey

World Bank Malawi high frequency phone survey between 2020 and 2024

  • Subset: 05/20-07/21
    • Only the first 13 rounds contain information about non-pharmaceutical mitigation behaviour
  • Nationally representative sample of households based on the Malawi National Integrated Household Panel Survey from 2019
  • 1729 participants at baseline & 8% attrition at R13
  • Further weighting based on census data

Why: Contributions

Emprical contribution

  • First causal investigation into health/income trade-off in context of protective measures during a pandemic
  • Qunatification of trade-off and comparison over time

Theoretical Contribution

  • Empirical test of widely-assumed trade-off assumption

Policy relevance

  • Informs more efficient data collections during early phases of emergencies
  • Informs ressource allocation efficiency

Chapter: Peer effects vs self selection